The Road to Safety: Network Effects from Collaboration
A while ago I wrote about the challenges for building maps and automotive AI, suggesting that the most sensible way forward relies on collaboration, sharing data, and computer vision. Yubin recently gave a good overview of how computer vision propels the collaborative approach to getting map data. Today, I’d like to dig a bit more into the collaboration part itself.
New York City (by mdroads)
To start with, I’d like to refer back to a recent post by Ben Evans about the winner-takes-all effects in autonomous cars. Ben suggests that there are network effects in collecting data to train pieces of the self-driving software stack as well as in collecting map data.
For example, the same object in different parts of the world can vary a lot in appearance. Also, the surroundings in which this object occurs, and hence, the background to recognize it against, varies a lot. The street scene in NYC is not the same as in New Delhi or São Paulo. Also, the scene is constantly changing in time, so there is a lot to be won from having very diverse data. This is what makes it hard to gather enough imagery to train your AI on your own. The more sources and the more diverse they are, the better your algorithms will be.
If you’re a small company, you don’t have the resources to send a fleet to every corner of the world. You might not have the resources to update your technology stack as often as other companies. But you do have resources for doing a part of it—and others like you will do other parts, which in combination will yield great results. What’s needed is a platform that enables automotive players to collaborate on data collection and understanding.
Take this passage from Ben’s essay:
“Maps have network effects. When any autonomous car drives down a pre-mapped road, it is both comparing the road to the map and updating the map: every AV can also be a survey car. If you have sold 500,000 AVs and someone else has only sold 10,000, your maps will be updated more often and be more accurate, and so your cars will have less chance of encountering something totally new and unexpected and getting confused. The more cars you sell the better all of your cars are—the definition of a network effect.”
Think about the very last sentence: the more cars you sell, the better all of your cars are. But in a non-monopoly situation, there will still be those other cars. All these other cars out there, that have less accurate, less frequently updated maps than yours, are going to have an effect on your car as well since they all operate in the same traffic environment.
It is, of course, better to own the car that is more capable of making sense of its surroundings. You will probably get to your destination faster and more efficiently compared to the other car that gets confused along the way. But what if that other car makes a mistake or ends up in an accident? How safe would you feel as a driver, a cyclist, or a pedestrian sharing the streets with that car?
If we talk about cars and traffic then we’re not just talking about economic loss or foregone profit—we’re talking about human lives. Choosing to work in silos instead of collaborating means holding back the development of traffic safety for everyone. (By the way, the aviation industry is trying to solve similar kinds of problems, related to passenger experience as well as aircraft safety.)
Ben also asks a final question: how much data do you really need? The need for data never ends, as the world is constantly changing. The reality is that no single actor alone can keep up with this, or even if they could, it would not be a sensible use of resources. Data should be contributed by everyone and available to everyone. If you want to compete, compete on what you do with the data downstream.
As a result of collaboration, our streets and roads will be safer, and innovation will increase as people find new creative uses for this information. From that perspective, there is no question whether to collaborate or not. The real question to think about is: what can we do with frequent, dense data capture of every city, available to all?